Provider review evidence

Minimal controls for reviewed API access.

This page summarizes the SENDFU provider API control plane currently available for reviewer inspection: account login, access review, hash-only API keys, monthly quota enforcement, usage accounting, and admin exports.

Public API surface

SENDFU exposes an OpenAI-compatible API for SENDFU-operated inference. SENDFU operates the public gateway and inference runtime as one service, with service credentials kept server-side. Request-bearing routes require a SENDFU-issued Bearer API key.

Catalog

GET /v1/models

Returns allowlisted model services that pass the current SENDFU availability check.

Inference

POST /v1/chat/completions

Accepts a selected model ID and processes authorized requests through the selected chat-completions service.

Status

GET /healthz

Reports application configuration and accounting-store checks. It does not by itself prove model inference availability.

Access workflow

Reviewer and customer access is separated from service credentials and is controlled through SENDFU accounts and reviewed API-key issuance.

Registration

/register creates a SENDFU account with email, company, and password.

/login establishes a signed SENDFU web session.

Access request

/dashboard/access-request records intended use, requested models, billing contact, and expected monthly tokens.

Admin review

/admin/access-requests allows an admin to approve or reject access requests.

API-key issuance

Approved requests receive a SENDFU inbound API key. The full key is shown once at issuance; the store keeps a hash and prefix, not the cleartext key.

API-key lifecycle and quota

The provider API uses inbound SENDFU keys. These are distinct from internal service credentials and can be independently limited or revoked.

Lifecycle

active / suspended / revoked

Only active API keys can call request-bearing routes. Admin pages support state changes for existing keys.

Model allowlist

per-key model control

Each key stores its allowed model IDs. Requests outside the key's allowlist are rejected before service processing.

Monthly quota

monthly_token_limit

Keys may have a monthly token ceiling. Over-limit requests are rejected with monthly_quota_exceeded.

Usage accounting and export

SENDFU records response usage when core token counters are present and rates each accepted request against the applicable published USD-per-token rate-card snapshot.

Recorded counters

Uncached prompt, cache-read, completion, and total token counts, together with status, model, request ID, user ID, API-key ID, billing period, rate-card version, applicable UTC pricing window, and rated USD amounts.

User view

/dashboard/usage shows recent usage and token totals for the signed-in account.

Admin JSON export

/admin/usage-export.json provides monthly rated usage summaries grouped for reviewed accounts and API keys.

Admin CSV export

/admin/usage-export.csv provides monthly rated usage rows suitable for reconciliation.

Rated invoice draft

/admin/invoice-draft.json returns a rate-card-backed monthly invoice draft for review. It is not evidence of payment, tax calculation, an issued final invoice, or a closed accounting period.

Credential, operator, and model boundary

The controls below are the relevant service boundary for provider review.

Boundary statement. Clients call SENDFU with SENDFU-issued inbound API keys. SENDFU does not accept client-supplied third-party keys. SENDFU operates the gateway, inference runtime, and deployed model services. Internal endpoints and credentials remain server-side. Model trademarks, developer identity, and base-model ownership remain with their respective owners unless explicitly stated otherwise.

Reviewer smoke-test shape

Use a reviewer-issued SENDFU API key. Do not send third-party credentials to SENDFU client endpoints.

curl https://sendfuglobal.com/v1/models

curl https://sendfuglobal.com/v1/chat/completions \
  -H 'Authorization: Bearer [SENDFU_REVIEWER_KEY]' \
  -H 'Content-Type: application/json' \
  -d '{
    "model": "z-ai/glm-5.1",
    "messages": [{"role": "user", "content": "Reply with one short sentence."}]
  }'

Streaming can be requested with "stream": true. Usage recording depends on service responses containing core token counters.